Colocating tasks in data centers using a side-effects performance model

Abstract

In data centers, many tasks (services, virtual machines or computational jobs) share a single physical machine. We explore a new resource management model for such colocation. Our model uses two parameters of a task-its size and its type-to characterize how a task influences the performance of the other tasks allocated on the same machine. As typically a data center hosts many similar, recurring tasks (e.g. a webserver, a database, a CPU-intensive computation), the resource manager should be able to construct these types and their performance interactions. In particular, we minimize the total cost in a model in which each task's cost is a function of the total sizes of tasks allocated on the same machine (each type is counted separately). We show that for a linear cost function the problem is strongly NP-hard, but polynomially-solvable in some particular cases. We propose an algorithm polynomial in the number of tasks (but exponential in the number of types and machines) and another algorithm polynomial in the number of tasks and machines (but exponential in the number of types and admissible sizes of tasks). We also propose a polynomial time approximation algorithm, and, in the case of a single type, a polynomial time exact algorithm. For convex costs, we prove that, even for a single type, the problem becomes NP-hard, and we propose an approximation algorithm. We experimentally verify our algorithms on instances derived from a real-world data center trace. While the exact algorithms are infeasible for large instances, the approximations and heuristics deliver reasonable performance.

Optimizing Egalitarian Performance when Colocating Tasks with Types for Cloud Data Center Resource Management

“…We generate a random sample of 10, 000 task records (see [27] for more information on the sample). Each task record corresponds to a task in our model.…”

“…In contrast, in our side-effects model [26], [27] rather than trying to predict tasks' performance from OS-level metrics, or ignoring it, we derive it from two characteristics of each task: task's type (e.g. : a database, or a computationally-intensive job) and task's size relative to other tasks of the same type (e.g.…”

“…In this paper, we study the egalitarian objective -we minimize the maximum of the tasks' costs (in our previous works [26], [27] we studied the utilitarian objective -we minimized the sum of the tasks' costs). The egalitarian objective corresponds to maximizing the quality of service proposed to all tasks.…”

“…Since the tasks are heterogeneous, the resulting performance dependencies are complex. In our previous work [26], [27] we proposed a new combinatorial optimization model that uses two parameters of a task -its size and its type -to characterize how a task influences the performance of other tasks allocated to the same machine.In this paper, we study the egalitarian optimization goal: the aim is to optimize the performance of the worst-off task. This problem generalizes the classic makespan minimization on multiple processors (P||C max ).…”

“…Since the tasks are heterogeneous, the resulting performance dependencies are complex. In our previous work [26], [27] we proposed a new combinatorial optimization model that uses two parameters of a task -its size and its type -to characterize how a task influences the performance of other tasks allocated to the same machine.…”